Education and training for preventing sharps injuries and splash exposures in healthcare workers

Summary of findings 1. Educational intervention compared to no intervention in healthcare workers for preventing sharps injuries and splash exposures

Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Educational intervention compared to no intervention in healthcare workers in cluster RCT
Patient or population: Healthcare workers Setting: Healthcare organisations Intervention: Educational intervention Comparison: No intervention
Outcomes	Risk with no intervention	Risk with Educational intervention	Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Sharps injuries, follow‐up 2 mo (CCT)	463 per 1 000	315 per 1 000 (222 to 440)	RR 0.68 (0.48 to 0.95)	395 (3 Controlled Clinical Trials)	⊕⊝⊝⊝ VERY LOW ^{1 2}
Needle stick injuries, questionnaires, 6 mo (RCT)	140 per 1 000	74 per 1 000 (29 to 188)	RR 0.53 (0.21 to 1.35)	167 (1 cluster‐RCT)	⊕⊕⊝⊝ LOW ^{1 2}
Needle stick injuries, hospital registers, 6 mo (RCT)	38 per 1 000	34 per 1 000 (14 to 83)	RR 0.91 (0.38 to 2.20)	529 (1 cluster‐RCT)	⊕⊕⊝⊝ LOW ³
Needle stick injuries, questionnaires, 12 mo (RCT)	119 per 1 000	49 per 1 000 (17 to 144)	RR 0.41 (0.14 to 1.21)	168 (1 cluster‐RCT)	⊕⊕⊝⊝ LOW ^{1 2}
Needle stick injuries, hospital registers, 12 mo (RCT)	41 per 1 000	19 per 1 000 (7 to 54)	RR 0.46 (0.16 to 1.30)	529 (1 cluster‐RCT)	⊕⊕⊝⊝ LOW ³
Injury rate (ITS), change immediately after intervention, low injury rate	The mean level of injury rates before was 35.9	MD 9.4 lower (14.9lower to 3.8 higher)	‐	2104 (2 Interrupted time series)	⊕⊝⊝⊝ VERY LOW ^{4 5}
Injury rate (ITS), change immediately after intervention, high injury rate	The mean level of injury rates before was 261.1	MD 77.1 lower (117.2 lower to 37.1 higher)	‐	255 (1 Interrupted time series)	⊕⊝⊝⊝ VERY LOW ^{4 5}
Injury rate (ITS), change in trend before and after intervention, low injury rate	The mean trend in injury rates before was ‐10.8	MD 2.3 lower (12.4 lower to 7.8 higher)	‐	2104 (2 Interrupted time series)	⊕⊝⊝⊝ VERY LOW ^{4 5}
Injury rate (ITS), change in trend before and after intervention, high injury rate	The mean trend in injury rates before was 9.6	MD 32.5 lower (49.6 lower to 15.4 higher)	‐	255 (1 Interrupted time series)	⊕⊝⊝⊝ VERY LOW ^{4 5}
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk ratio; MD: Mean Difference
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹ We rated down one level because of imprecision based on wide confidence intervals. ² We rated down with two levels because of risk of bias due to self‐reporting and missing data ³ We rated down with two levels because of wide confidence intervals. ⁴ We rated down one level because of risk of bias due to missing period for data points and few data points. ⁵ We rated down one level because of risk of bias due to a decreasing pre‐intervention trend of injuries in all studies (intervention effect was difficult to detect).

Background

Description of the condition

In healthcare settings, workers are at risk of acquiring infectious diseases through sharps injuries and splash exposures to blood or bodily fluids. The blood‐borne viruses with the most severe consequences include hepatitis B (HBV), hepatitis C (HCV) and the human immunodeficiency virus (HIV) (Pruss‐Ustun 2005). Worldwide, healthcare workers (HCWs) experience over 3 million sharps injuries each year (Pruss‐Ustun 2005). This leads to approximately 66,000 HBV infections, 16,000 HCV infections and 1000 cases of HIV annually (Pruss‐Ustun 2005). Developing countries experience a high prevalence of HIV, and because of limited resources for prevention, there is a high rate of needle stick injuries (Enwere 2014). Exposure injuries among HCWs are also commonly not monitored in developing countries (Pruss‐Ustun 2005). The risk of potentially experiencing a sharps injury at work and the concern of acquiring an infectious disease as a direct result of this exposure may cause psychological stress for the HCWs. In turn, this stress impacts on both the workers' functioning in their work and personal life (Fisman 2002; Sohn 2006). There is also the financial burden imposed on employers due to occupational exposure to blood‐borne infectious diseases, which includes costs related to blood tests, treatment, outpatient visits and lost working hours (Lavoie 2014). The actual causes of sharps injuries are multifactorial and include elements that could be modified by educational interventions. Such causes include incorrect use of personal protective equipment, professional inexperience, a lack of education and training on infection control and occupational health principles, improper management of sharps, and a subjective perception of risk (Akduman 1999; Ansa 2002; Clarke 2002; Doebbeling 2003; Fisman 2007; Ilhan 2006; Oh 2005; Orji 2002; Roberts 1999; Smith 2006a; Smith 2006b; Wallis 2007).

Description of the intervention

The focus of this review was education and training interventions that aimed to prevent sharps injuries and splash exposures among workers within healthcare settings, through the elimination or management of hazards in the workplace and changes in healthcare workers' skills and behaviors (Verbeek 2004). Educational interventions aim to decrease sharps injuries and splash exposures by increasing HCWs' knowledge, behavior or processes regarding the correct choice and safe operation of needles, scalpels and other sharp devices necessary in the delivery of healthcare. We defined education as the imparting or shared creation of knowledge, and we considered workers' education to be comprised of their knowledge, attitudes and skills (Verbeek 2004). Educational interventions may consist of group‐based instruction or other types of information delivery such as videos, leaflets, protocols, and guidelines given to people to watch or read in their own time. Educational meetings can be either didactic (e.g. lecture presentations) or interactive (e.g. workshop with role‐play and case discussions). However, repeated active educational interventions that promote interactivity have a higher chance of altering the behavior of HCWs and sustaining such changes (NHMRC 2010). The intervention can be delivered by an Occupational Safety and Health department, hospital infection control committee, or another educational body. We defined training as the imparting or shared practice of skills. Training interventions vary in mechanisms used in different healthcare settings. Common approaches include self‐directed learning modules, presentations, education delivered by a visiting expert, inter‐professional education, interactive web‐based training, and seminars.

How the intervention might work

The key to preventing sharps injuries and splash exposures is HCWs' ability to identify risky situations, react accordingly, act in a safe manner, and consistently and correctly handle equipment. Understanding the mechanisms of injury may also help HCWs to identify risky situations. Training refers to planned efforts to facilitate the learning of specific competencies (Noe 2005; Robson 2010). If underpinned by specialised skills and knowledge, these competencies can help to guide changes in HCWs' behavior (Robson 2010). There are different levels of engagement in training, and we employed the classifications used by Burke and Robson (Burke 2006; Robson 2010). Low‐engagement training employs oral, written or multimedia presentations by an expert and requires no active participation by the worker. Medium‐engagement training employs a greater degree of worker involvement, such as lectures with a group discussion and quizzes with feedback. Finally, in high‐engagement training, the worker plays an active role in the training activities through face‐to‐face or virtual settings and hands‐on practice opportunities (Burke 2006). There are several ways of decreasing or removing exposure to blood and bodily fluids, including the removal of hazards at the source such as the abolition of unnecessary injections, or along the care pathway by using safety medical devices, instituting safer workplace practices or using personal protective equipment (Ellenbecker 1996; Roelofs 2003). The Center for Disease Control and Prevention have also determined that sharps injuries generally occur during a procedure on a patient, after use, and before, during or after the disposal of a sharp instrument (CDC 2008).

Why it is important to do this review

Education and training for the prevention of sharps injuries and splash exposures are part of obligatory professional development for HCWs in certain countries. Since practices vary and there are also legislative requirements to protect workers' health and safety, it is important to study the effects of education and training to prevent occupational sharps injuries and splash exposures by means of a Cochrane review. This review is one of a group of Cochrane reviews that address interventions to prevent percutaneous exposure injuries. The other three Cochrane reviews have assessed the effectiveness of blunt needles (Parantainen 2011), safety devices (Lavoie 2014; Reddy 2017), and gloves (Mischke 2014). Parantainen 2011 found high‐quality evidence showing that the use of blunt needles appreciably reduced the risk of exposure to blood and bodily fluids for surgeons and their assistants over a range of operations. In Lavoie 2014, there was only very low‐quality, inconsistent evidence that most safety‐engineered devices (SEDs) prevented sharps injuries. The risk of blood contamination was greater with devices that had to be actively switched on. This review was updated in 2017 (Reddy 2017) and determined that there was still low‐quality and inconsistent evidence on safety‐engineered devices preventing sharps injuries. Alternatively, Mischke 2014 found moderate‐quality evidence that double‐gloving, when compared to single‐gloving during surgery, reduces perforations and bloodstains, indicating a reduction in sharps injuries.

In addition to Cochrane reviews, there are other published systematic reviews that have assessed interventions to prevent sharps injuries. A systematic review from 2015 focused on educational interventions alone or in combination with safety‐engineered devices (Tarigan 2015). The review's meta‐analysis utilizing multiple study designs showed that training decreased the rate of sharps injuries by 34%, while SEDs were responsible for a 49% reduction. However, the combination of training interventions with SEDs led to a 62% drop in the rate of sharps injuries. However, there was only one randomized controlled trial (RCT) included in the review. A recent systematic review investigating the effects of training and SEDs on needle‐stick injury rates in European countries identified six studies that demonstrated a reduction in injury rates due to the combination of training and SEDs (Aziz 2018). The aim of this Cochrane review was to evaluate a broader range of controlled trials of various types of education and training on the incidence of sharps injuries and splash exposures in HCWs.

Objectives

To assess the effects of education and training interventions compared to no intervention or alternative interventions for preventing sharps injuries and splash exposures in healthcare workers.

Methods

Criteria for considering studies for this review

Types of studies

All published and unpublished randomized controlled trials, cluster‐randomized controlled trials (cluster‐RCTs), and controlled before‐and‐after (CBA) studies including interrupted time series (ITS) were considered for inclusion in this review. We employed the Cochrane Effective Practice and Organisation of Care Review Group's (EPOC 2013) definition of an interrupted time series design as at least three outcome measurements before and at least three outcome measurements after the implementation of an intervention. We included studies irrespective of language of publication, publication status, publication date or the use of blinding.

Types of participants

We included workers in all healthcare settings, which meant paid employees aged 18 or over who were professionally involved in healthcare settings and who were at risk of sharps injuries or splash exposures to blood or bodily fluids. Over half of the study participants had to fulfil this criterion for the study to be included in the review. Studies on nursing and medical students were excluded as the participants were not paid employees in a healthcare facility and the interventions occurred as a component of their tertiary studies.

Types of interventions

Inclusion criteria

We included all educational and training interventions that provided information to HCWs with the aim to change knowledge, behaviors and processes related to sharps injuries and splash exposure occurrence. Studies that had education or training as an intervention were included. Given the difficulties with identifying the effect of education and training in co‐intervention studies, it was decided to exclude studies with safety‐engineered devices (SEDs) or sharps‐disposal containers as a co‐intervention. This also avoids overlap with another Cochrane review in our series on SEDs by Reddy and colleagues (Reddy 2017).

Exclusion criteria

We considered studies where education and/or training was the sole intervention, therefore studies with co‐interventions were excluded from this review. As the other Cochrane reviews in our series have covered blunt needles, SEDs and gloves, studies employing these interventions were excluded.

Types of outcome measures

We included studies that reported on the following primary outcomes.

Primary outcomes

Studies that evaluated the effectiveness of interventions on incidence of sharps injuries and splash exposures in healthcare settings were included. The primary outcomes were:

sharps injuries (eg. needle stick injuries)
splash exposure (eg. bodily fluid)

This exposure was measured by:

self‐reported questionnaires;
reporting of sharps injuries and splash exposures through employer records;
or empirical observations by researchers.

Secondary outcomes

We considered the following five items as secondary outcomes:

observed cases of infections;
knowledge related to sharps injuries and splash exposures as measured by surveys to HCWs;
knowledge related to infectious disease transmission as measured by surveys to HCWs;
adherence to organisations' occupational health and safety policies and procedures, as observed or reported;
behavioral changes in clinical practice, as observed or reported.

Search methods for identification of studies

Electronic searches

We applied search terms for sharps injuries and splash exposures and combined these with the recommended search strings for randomized trails and non‐randomized studies. We employed the search strategy from (Robinson 2002) to guide the search for randomized clinical trials and controlled clinical studies. To identify non‐randomized studies, we used the sensitive search strategy for occupational health intervention studies (Verbeek 2005). We used these strategies to search CENTRAL, OVID MEDLINE, Embase, NHSEED, Science Citation Index Expanded, CINAHL and OSH‐update. In addition, we searched the databases of Global Health, AustHealth and Web of Science. The original search strategy is outlined in (Appendix 1) and was run in February 2016. The original search strategy was re‐run in November 2019, and again in February 2020. In April 2020, the search strategy was updated and run in CINAHL, OVID MEDLINE, Scopus and Web of Science, as illustrated in (Appendix 2). The updated searches were done at the library of University of Helsinki, which did not have a licence to Embase. Hence, that search was not re‐run.

Searching other resources

We searched the reference lists of all studies in full‐text screening for additional relevant studies. The reference lists of all related systematic reviews were also searched. When required details of the included studies were absent, we contacted the first authors for further information. Additional data were provided by (El Beltagy 2012) and colleagues; additional information was also sought from (Bijani 2018), however, this was not provided.

Data collection and analysis

Selection of studies

Using the inclusion and exclusion criteria, two authors (SC, HL) worked independently to screen the titles and abstracts of the references yielded by the search strategy using online software (Covidence). Full‐text versions of the references that met the inclusion criteria were obtained. Disagreements regarding the inclusion of certain studies were solved by discussion and review with a third review author (JL).

Data extraction and management

Two researchers (SC, HL) independently used a specifically created data extraction form to extract the essential study characteristics for each study. These characteristics included information related to study design, participants, intervention, outcome and results.

Assessment of risk of bias in included studies

Two review authors (HL, JL) assessed the risk of bias independently for each included study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). As blinding is unfeasible for educational interventions, we assessed the following items to measure the risk of bias in included studies: randomization, allocation concealment, incomplete outcome data and selective outcome reporting. Deviations from the protocol are reported in the (Differences between protocol and review) section of this review. For ITS studies, we used the 'Risk of bias' criteria as presented by (Ramsay 2003).

Measures of treatment effect

Because in the CBA studies, the baseline rates were high, we used risk ratios (RRs) for dichotomous outcomes instead of odds ratios (ORs) as the measure of the education and training effect on sharps injuries. For knowledge and behavior changes, we used mean differences in the measurement scales. For ITS studies, we extracted and re‐analyzed the data from the original papers according to the recommended methods for the analysis of ITS designs for inclusion in systematic reviews (Ramsay 2003). These methods utilize a segmented time‐series regression analysis to estimate the effect of intervention while taking into account secular time trends and any auto‐correlation between observations. For each study, we fitted a first‐order autoregressive time‐series model to the data using a modification of the parametrization of (Ramsay 2003). Details of the model specification are as follows:

Y_t = β₀ + β₁Time_t + β₂ (Intervention_t ) + β₃ (Time after individual intervention_t ) + e_t

Y_t is the outcome at time ‘t’;

‘Time’ indicates the number of time units (i.e. years, in our case) from the start of the series;

‘Intervention’ is a dummy variable taking the values ‘0’ for pre‐intervention and ‘1’ for the post‐intervention segment;

‘Time after intervention’ is taking values ‘0’ in the pre‐intervention and counts the time units (i.e. ‘years’) in the post‐intervention segment at time ‘t’;

β₀ estimates the base level of the outcome at the beginning of the series (at time ‘t’ = 0);

β₁ estimates the change in outcome per time unit (i.e. ‘year’) in the pre‐intervention segment;

β₂ estimates the change in (short‐term) level in the post‐intervention segment;

β₃ estimates the change in (long‐term) trend in the post‐intervention segment; and

e_t estimates the error.

We used the change in level and change in slope as measures of short‐term and long‐term treatment effects, respectively, for ITS studies. Individual studies parameter estimates were combined and weighted by the inverse variance method, in accordance with Cochrane recommended methodology (Borenstein 2010). ITS analysis was performed using Statistical Analysis System version 9.4 (SAS 2013). A random‐effects meta‐analysis of ITS findings was conducted manually, following the formulae provided in (Borenstein 2010).

Although three studies reported having used the ITS approach in their data analysis and reporting, none of them reported the findings in accordance with this methodology. As such, the first step required in the statistical analysis of the current review was to conduct the ITS analysis on the data of each of the three studies separately. Due to limited data time points available for one study (Zafar 2009), ITS analyses were conducted with respect to 'pre‐intervention' and 'post‐intervention' periods only. That is, where data existed for the during‐intervention period, these were treated as 'post‐intervention'.

Data for each of the three ITS studies were first analyzed separately, using the respective study’s original scale or metrics. Parameter estimates for one study (Sarbaz 2017) were then converted to yearly estimates, before being combined with the ITS results of two other studies (El Beltagy 2012; Zafar 2009) using yearly data for meta‐analysis.

Unit of analysis issues

For studies that employed a cluster‐randomized design but did not make an allowance for the design effect, we aimed to calculate the design effect by the methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) for the calculations. However, the only cluster‐RCT included did not report the size of clusters and also had a 50% loss of follow‐up (Van der Molen 2011).

Dealing with missing data

We contacted the authors for additional information, if required. If the data were presented in figures only and we did not receive a response from the authors concerned, we extracted the available data from the figures presented in the article. If data such as standard deviations were missing and could be calculated from other data present in the article, such as P values, we did so according to the recommendations in Higgins 2011.

Assessment of heterogeneity

We defined studies to be sufficiently homogeneous when they had similar populations, interventions and outcomes measured at the same follow‐up time point. We considered all healthcare workers as sufficiently similar to assume a similar preventive effect from education and training. We divided outcomes into number of exposures, number of infections, and changes in knowledge or behavior. We deemed all educational and training interventions within these categories to be conceptually similar and sufficiently homogeneous to be combined in a meta‐analysis. We did not combine various study designs, as we assumed that there were large differences in the risk of bias between different study types. Hence, we presented the results per comparison separately per design type. We assessed statistical heterogeneity by means of the I² statistic. We used the values of < 40%, between 30% and 60%, between 50% and 90%, and over 75% as indicating not important, moderate, substantial and considerable heterogeneity, respectively, as proposed in the Cochrane handbook (Higgins 2011).

Assessment of reporting biases

We aimed to assess publication bias with a funnel plot if more than five suitable studies were available for inclusion in the plot. However, there were fewer than five studies in each of the result groups.

Data synthesis

We pooled studies with sufficient data that we judged to be homogeneous by design and statistically speaking, using RevMan software (RevMan 2014). We used a random‐effects model since the studies were heterogeneous by their design and their participant populations. Finally, we used the GRADE approach to assess the quality of the evidence per comparison and per outcome as described in the Cochrane handbook (Higgins 2011). We reduced the quality of evidence by one or more levels if there were limitations in one or more of the following domains: risk of bias, consistency, directness of the evidence, precision of the pooled estimates, and the possibility of publication bias. For the two most important outcomes, the number of exposures and the number of observed cases of infections, we used the GRADEproGDT programme to generate 'Summary of findings' tables (GRADEproGDT 2015).

Subgroup analysis and investigation of heterogeneity

We aimed to analyze whether there was a difference in the intervention effect in workers with different levels of exposure, which we would categorize as low or high exposure. However, we were not able to do separate analyses for these subgroups because we did not have sufficient studies to make this comparison. We also aimed to classify the educational interventions according to their level of engagement to either a low or high level of engagement. For this subgroup analysis, we considered both medium and high‐intensity training to constitute a high‐engagement intervention. However, there were no studies of the low level of engagement and we were unable to undertake this analysis.

Sensitivity analysis

For a sensitivity analysis, we aimed to analyze the results again, including only studies we judged to have a low risk of bias. We categorized studies based on their design as described above in (Types of studies). However, all studies had a high risk of bias and we were unable to undertake this analysis.

Summary of findings and assessment of the certainty of the evidence

We assessed the certainty of the evidence for the main outcomes using the GRADEpro software. The 'Summary of findings' (SoF) tables are attached to the review (summary of findings Table 1).

Results

Description of studies

Results of the search

Figure 1 illustrates the study flow diagram for the electronic searches. For the original search, conducted in February 2016 and updated in July 2016, we found 6319 studies. After duplicates were removed, the titles and abstracts of 5817 studies were assessed in accordance with the inclusion criteria using online software (Covidence). The original searches were re‐run in November 2019 and in February 2020, with a further 474 studies added to Covidence for screening. In April 2020, the searches were updated and re‐run, resulting in another 66 studies added to title and abstract screening. As a result of this screening, 6271 studies were found to not have met the criteria and were excluded, leaving 86 studies to proceed to full‐text assessment. Of these 86 studies, seven were deemed to have met the inclusion criteria while 79 were excluded; please see the Characteristics of excluded studies table for reasons for exclusion.

Figure 1

Study flow diagram

Included studies

We included seven studies in this review; please see the Characteristics of included studies table for further information.

Types of study designs

The following study designs were utilized to assess the effects of the intervention:

one cluster‐randomized controlled trial (cluster‐RCT) (Van der Molen 2011);
three controlled clinical trials (CCTs) (Bijani 2018; Choi 2009; Huang 2002);
three interrupted time series (ITS) studies (El Beltagy 2012; Sarbaz 2017; Zafar 2009).

Geographical location

The included studies originated from six different countries. Two of these studies were from Iran (Bijani 2018; Sarbaz 2017), with one article originating from each of the following countries: the Netherlands (Van der Molen 2011), South Korea (Choi 2009), China (Huang 2002), Saudi Arabia (El Beltagy 2012), and Pakistan (Zafar 2009).

The baseline rate of sharps injuries

The baseline rate of sharp injuries varied between the cluster‐RCT, the ITS studies, and the controlled clinical trials.

In studies with hospital registry systems, the baseline rates of injuries varied from 43 to 203 injuries per 1000 HCWs per year; in the cluster‐RCT 68 per 1000 HCW per year (Van der Molen 2011) and in the ITS studies 43 per 1000 HCW per year (Zafar 2009); 38 per 1000 HCW per year (El Beltagy 2012); and 203 per 1000 HCW per year (Sarbaz 2017). In these studies, the observations were surveillance‐based.

In questionnaire‐based studies with shorter follow‐up periods, the rates of injuries were higher: from 1800 to 7000 injuries per 1000 HCWs per year. The calculated yearly incidence was: 3300 per 1000 HCW (Choi 2009), 2100 per 1000 HCW (Bijani 2018) and 588 per 1000 HCW (Huang 2002).

In the only study that reported both questionnaire‐based results and registry‐based data, the rates were higher in the questionnaire‐based results (Van der Molen 2011).

Year of publication

All studies were published after the year 2000 (Bijani 2018; Choi 2009; El Beltagy 2012; Huang 2002; Sarbaz 2017; Van der Molen 2011; Zafar 2009).

Type of setting

All studies were conducted in a hospital setting (Bijani 2018; Choi 2009; El Beltagy 2012; Huang 2002; Sarbaz 2017; Van der Molen 2011; Zafar 2009).

Participants

The authors of three studies used the broad term of healthcare workers to describe their participants (El Beltagy 2012; Van der Molen 2011; Zafar 2009). The category was expanded on to include such occupational groups as physicians, nurses, respiratory therapists, phlebotomists, medical technologists, midwives, technicians, housekeeping professionals, and others. HCWs were defined as physicians, nurses, and paramedics in one study (Sarbaz 2017). One study included physicians and nurses only (Van der Molen 2011). Three studies only included nurses as their participants (Bijani 2018; Choi 2009; Huang 2002).

Sample size and follow‐up measurements

The cluster‐RCT (Van der Molen 2011) identified 796 eligible HCWs that were allocated to either one of the two experimental groups (one education‐only group, the other education + safety devices group) or the control group, with approximately 265 participants allocated to each group. A total of 263 HCWs from eight hospital wards were allocated to the educational workshop‐only group. Of the 796 enrolled participants, only 160 completed a questionnaire for all three study time points, at baseline, six, and twelve months follow‐up. For the three controlled clinical trials, Bijani 2018 determined that a sample size of 120 participants, 60 nurses per group, was sufficient and noted that all participants completed the two months follow‐up questionnaire. Choi 2009 began with 110 participants in both the control and experimental groups and finished the study with 92 and 85 participants respectively at two months follow‐up. Huang 2002 enrolled 100 participants equally divided between the control and experimental groups. One participant from each group withdrew from the study resulting in 49 participants in each of the control and experimental groups at four‐month follow‐up.

Of the ITS studies, El Beltagy 2012 stated that in the pre‐intervention period, from 1997 to 2000, there were 11,093 HCWs in their institution, no staff numbers were provided for the intervention years of 2001 to 2003 (resulting in this period being excluded from our analyses), while 25,027 HCWs were reported at the institution over the post‐intervention period of 2004 to 2008. For the purpose of the meta‐analysis, these numbers were taken as the (proxy) sample sizes. In Sarbaz 2017, the intervention hospital was comprised of 1884 total HCWs broken down as 1271 nurses, 214 physicians, and 399 paramedics, while the control hospital had a total of 1875 HCWs comprised of 1206 nurses, 371 physicians, and 298 paramedics over the four‐month post‐intervention follow‐up period. Data on the control hospital were not used in the meta‐analysis. Of the three years, in the pre‐intervention period, Zafar 2009 reported 4788 HCWs at their institution in 2002, 5261 HCWs in 2003, and 5861 HCWs in 2004. There were 6566 HCWs at the institution for the intervention year 2005. For the two post‐intervention years, HCW numbers were reported as 7635 in 2006 and 7468 in 2007.

Interventions

The mode of delivery of the education and training interventions varied from study to study, as illustrated in (Table 1). The educational workshop in Van der Molen 2011 was delivered via an interactive presentation. Bijani 2018 employed such educational strategies as lectures, videos, question‐and‐answer sessions, and demonstrations. A web‐based educational program was introduced by Choi 2009, which tested participants' knowledge before and after completing five modules. Huang 2002 utilized two different approaches, small group discussions were used to clarify the risks associated with blood‐borne viruses, while demonstrations focused on universal precautions including the use of personal protective equipment, hand‐washing, handling of sharps, correct techniques for common procedures, and sharps disposal containers. Printed materials, including pamphlets, slides, and photographs were also utilized in Huang 2002. El Beltagy 2012 utilized educational presentations, including an interactive practical session on the correct use of sharps, PI (percutaneous injuries) prevention week, poster and flyer competitions, and a PI message of the month was sent to the staff via email. Sarbaz 2017 and colleagues utilized a web‐based information system. Zafar 2009 used workshops, poster and flyer competitions; an infection control message of the month was also sent to staff via email to deliver their educational intervention.

Table 1. Interventions from included studies

Author, Year, Study Design	Educational Presentation	Interactive Demonstration	Marketing Tools	Web‐Based Education
Bijani 2018, CCT	X	X
Choi 2009, CCT				X
El Beltagy 2012, ITS	X	X	X
Huang 2002, CCT	X	X	X
Sarbaz 2017, ITS				X
Van der Molen 2011, cluster‐RCT	X	X
Zafar 2009, ITS	X	X	X

In Huang 2002, three investigators trained in universal precautions from the Yale School of Nursing delivered the educational intervention. The hospital's infection control department delivered the educational presentation for staff in El Beltagy 2012. Zafar 2009 advised that the hospitals' infection control committee conducted the educational intervention. The personnel who delivered the education and training interventions was not stated in four trials (Bijani 2018; Choi 2009; Sarbaz 2017; Van der Molen 2011).

In Van der Molen 2011, participants discussed the cause, prevalence, and strategies for the prevention of exposure injuries within their hospital. The content of the intervention reported by Bijani 2018 included standard precautions, management, and reporting of an injury. The topics of the five educational models in Choi 2009 were the epidemiology of blood‐borne infections, statistical data on blood‐borne infections, the practice of just‐in‐time exposure (including a link to report the exposure), post‐exposure practice, and prevention strategies. Huang 2002 focused on blood‐borne viruses and universal precautions, with a specific focus on the risk of transmission of blood‐borne viruses, injury reporting, and management and prevention strategies for staff to employ in their professional practice. El Beltagy 2012 focused their educational program on the prevention of exposure injuries, with a particular focus on risk factors associated with injury, needles and devices, risk of transmission of blood‐borne viruses, first aid, and the immune status of employees. The education in Sarbaz 2017 was on infection control and prevention. Zafar 2009 chose to focus their educational intervention on safe workplace practices including standard precautions such as appropriate personal protective equipment (PPE) use, hand hygiene, and safe sharps disposal.

The one‐hour workshop in Van der Molen 2011 was delivered during the change over between morning and afternoon shift and occurred two to three times over four months. The continuing educational program introduced by Bijani 2018 lasted for 10 hours. The HCWs in Choi 2009 accessed the web‐based educational program for 30 minutes per week over three weeks, 90 minutes in total. The intervention in Huang 2002 was delivered in a two‐hour lecture, a one‐hour demonstration, and a 30‐minute discussion. In El Beltagy 2012, new employees underwent a two‐hour educational presentation, while onboard staff received regular one‐hour presentations, with the whole hospital receiving a refresher course annually. The duration of the program was not specifically mentioned in Sarbaz 2017, however, the importance of the system was mentioned at monthly infection control sessions and morning sessions over three days. Duration of the educational intervention was also not mentioned in Zafar 2009.

Levels of participant engagement in training as outlined by Burke and Robson (Burke 2006; Robson 2010) can be viewed in Table 2. One study (Van der Molen 2011) was determined to have a medium level of engagement due to the active engagement of participants. Another study (Bijani 2018) had a question‐and‐answer session; this was seen as an example of medium engagement. The web‐based education program in one study (Choi 2009) had pre‐ and post‐assessments for each of the five modules, and this was seen as medium engagement with participants. One study (Huang 2002) utilized a 30‐minute discussion and this was determined to be medium engagement. Another study (El Beltagy 2012) employed interactive practical sessions on the correct usage of sharps, in order to reinforce correct and safe practices and this was seen as a high level of participant engagement in training. Little detail was provided on the web‐based educational program in one study (Sarbaz 2017), therefore, the level of participant engagement could not be determined. One study (Zafar 2009) utilized interaction sessions with staff, and this was seen as medium engagement.

Table 2. Level of participant engagement in training all studies

Author, Year, Study Design	Low Level of Engagement	Medium Level of Engagement	High Level of Engagement	Unable to Identify
Bijani 2018, CCT		X
Choi 2009, CCT		X
El Beltagy 2012, ITS			X
Huang 2002, CCT		X
Sarbaz 2017, ITS				X
Van der Molen 2011, cluster‐RCT		X
Zafar 2009, ITS		X

Primary Outcome

All of the included studies addressed our primary outcome, the exposure of workers to blood or bodily fluids in healthcare settings as measured by the incidence of sharps injuries and splash exposures. However, the splash exposures were reported only in two studies (Sarbaz 2017, Bijani 2018), and only in (Sarbaz 2017) separately from sharps injuries. Authors in one study (Van der Molen 2011) used questionnaires and hospital registry data to identify the incidence of sharps injuries; they also evaluated behavioral measures. In another study (Bijani 2018), the incidence of sharps injuries were measured by questionnaire. Authors of another study (Choi 2009) reported on secondary outcomes and the incidence rate of sharps injuries and this was calculated by the number of injuries/number of nurses x 100. The data were collected by questionnaire. Huang 2002 used a questionnaire to collect data on sharps injuries. El Beltagy 2012 examined sharps injury rates by using data from the hospitals' surveillance system. Injury rates were calculated by the number of reported injuries per year/the number of HCWs per year. Zafar 2009 used surveillance data to collect the incidence rate of sharps injuries, which was calculated as needle‐stick injuries (NSI) x 100 full‐time equivalent (FTE).

Secondary Outcomes

The secondary outcomes were reported by the included studies as follows; 1) observed cases of infections (none), 2) knowledge related to sharps injuries as measured by surveys to HCWs (Choi 2009; Van der Molen 2011), 3) knowledge related to infectious disease transmission as measured by surveys to HCWs (Choi 2009; Huang 2002), 4) adherence to organizations' occupational health and safety policies and procedures, as observed or reported (Huang 2002; Van der Molen 2011), 5) behavioral changes in clinical practice, as observed or reported (El Beltagy 2012; Huang 2002; Van der Molen 2011; Zafar 2009). In order to assess the secondary outcomes, further information was requested from Bijani 2018, however, this was not provided and the assessment was unable to be conducted.

Excluded studies

The Characteristics of excluded studies section presents the reasons for exclusion for the 79 studies. Studies were excluded for having the wrong study design (Adams 2006; Al Maqbali 2014; Askari 1989; Askarian 2011; Aziz 2009; Aziz 2018; Beekmann 2001; Brusaferro 2009; Cantineau 2002; Chin 2001; Duerink 2006; Enwere 2014; Haiduven 1992; Haiduven 1995; Higginson 2013; Hiko 2012; Holodnick 2000; Hooper 2005; Hunt 2004; Linnemann 1991; Louis 2002; Marziale 2010; Mendelson 2003; Mobasherizadeh 2005; Moens 2004; Qaiser 2013; Rajkumari 2015; Regez 2002; Ribner 1990; Richard 2001; Rogers 2000; Roudot‐Thoraval 1999; Sellick 1991; Sossai 2010; Tarigan 2015; Trape‐Cardoso 2004; Tuma 2006; Visser 2006; Whitby 1991; Whitby 2002; Yang 2011; Zawilla 2013), not addressing the primary outcome (Abdo 2019; Ahmed Saleh 2017; Amuwo 2011; Gramling 2013; Krishnan 2007; Lueveswanji 2000; Nour‐Eldein 2016; Srikrajang 2005; Wright 1997), not implementing an education and training intervention (Adams 2003; Akbari 2018; Al‐Zahrani 2014; Amira 2014; Ashat 2011; Azar‐Cavanagh 2007; Bodkin 2003; Fathi 2017; Fritzsche 2016; Fukuda 2016; Hagstrom 2006), wrong participants (Amuwo 2013; Askarian 2006; Calabro 1998; Dante 2014; Froom 1998; Seng 2013; Wang 2003; Wu 2020; Yang 2007; Zhang 2010; Zhang 2013), and wrong intervention (Apisarnthanarak 2008; Gershon 1999; Kanamori 2016; Stringer 2009; Valls 2007; Zafar 1997).

Risk of bias in included studies

We presented the assessment of the 'Risk of bias', in the 'Characteristics of included studies' table (Characteristics of included studies). We then summarized the results of the risk of bias per study in the (Risk of bias in included studies) section (Figure 2) (Figure 3). The 'Risk of bias' table contains the risk of bias assessments for both randomized and non‐randomized studies, and hence not all cells were applicable to both of these study types and thus some of these remained unclear. Since the bulk of the evidence came from the ITS studies, we presented the risk of bias for these studies in a separate table (Table 3). We judged all included studies to have a high overall risk of bias. Random allocation was performed only by the cluster‐randomized trial (Van der Molen 2011). Selection bias was another major challenge, as the selection of wards was often unclear and there were sometimes differences at baseline in the characteristics of intervention and control wards.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

See: Summary of findings 1 Educational intervention compared to no intervention in healthcare workers for preventing sharps injuries and splash exposures

Table 3. Risk of bias in the ITS studies

Author, Year, Study Design

Intervention independent from other changes

Sufficient data points

Test for trend

Intervention did not affect data collection

Blinded outcome assessment

Complete data set

Reliable outcome measures

Total

El Beltagy 2012, ITS

Done (1)

This was an

educational intervention only.

Done (1)

Four data points before and five data

points after the intervention

Done (1)

We re‐analyzed

the study for trend.

Done (1)

Data collection remained the same pre‐

and post‐

intervention.

Not clear (0)

The reporting of NSI was done by a net service and the protocol remained the same.

Done (1)

All hospital employees were included in the study.

Done (1)

Hospital had a sharp injury surveillance system prior to and after the intervention. Although not ideal as the possibility of under‐reporting, but appropriate for the study outcome.

Sarbaz 2017, ITS

Done (1)

An education

alone intervention

Done (1)

Four data points

before and four data points after the intervention

Done (1)

We re‐analyzed the study for trend.

Done (1)

Data collection remained the same pre‐ and post‐

intervention.

Done (1)

Infection personnel

and staff were blinded to the study.

Done (1)

All hospital employees were included in the study.

Done (1)

Hospital had a sharp injury surveillance system prior to and after the intervention. Although not ideal as the possibility of under‐reporting, but appropriate for the study outcome.

Zafar 2009

Zafar 2009, ITS

Done (1)

An education

alone intervention

Done (1)

Three data points before and three data points after the intervention

Done (1)

We re‐analyzed the study for

trend.

Done (1)

Comment:

Data collection remained the same

pre‐ and post‐

intervention.

Not clear (0)

Authors did not provide information on blinding.

Done (1)

All hospital employees were included in the study.

Done (1)

Hospital had a sharp injury surveillance system prior to and after the intervention. Although not ideal as the possibility of under‐reporting, but appropriate for the study outcome.

NSI: needle‐stick injury

Allocation

There was one cluster‐randomized controlled trial (Van der Molen 2011), which did not explicitly report the randomization process. Hence, we considered the risk of bias to be unclear. The controlled clinical trial (Huang 2002) from China took a random sample of 100 nurses in a hospital. They were randomized 1:1 to both the experimental and control groups (n = 50 in each group), with all the nurses agreeing to participate. Since the randomization process was not well described, we considered the risk of selection bias to be unclear. The body text of the study by (Choi 2009) was in Korean although the abstract and tables were in English. The details of selection were difficult to assess because we were unable to arrange translation of the study into English.

Blinding

In the cluster‐RCT (Van der Molen 2011), the outcomes were assessed by officially registered injuries. We judged the risk of bias to be low. In the controlled clinical trials (Choi 2009; Huang 2002), blinding was not described, therefore we considered the risk of these studies as unclear.

Incomplete outcome data

For the cluster‐RCT (Van der Molen 2011), we considered there was a high risk of attrition bias. At cluster level, all wards remained in the study but at the individual level, 49% were lost at 12 months follow‐up for the main outcome measure of self‐reported NSIs compared to baseline. In the controlled clinical trials (Choi 2009; Huang 2002), the risk of incomplete outcome data appeared to be low. In the (Huang 2002) study, 98% of nurses were included in the follow‐up. According to the tables in (Choi 2009), all HCWs remained at the follow‐up timelines.

Selective reporting

The cluster‐RCT (Van der Molen 2011) and two of the controlled clinical trials (Choi 2009; Huang 2002) reported all outcomes as described in the methods section. Hence, we rated the risk of bias as low. Also, in the ITS studies, hospital registers were used for reporting and we considered the risk of reporting bias was low.

Other potential sources of bias

For the non‐randomised trials (NRS), we observed two biases: confounding, whether there were other interventions that might have confounded the results and selection bias, whether there was bias in the method of selection ward to the study. Overall, we rated the risk of these biases low or uncertain.

There was very little information on the funding sources for the included studies. See (Table 3) for an overview of our assessment of the three ITS studies in all seven risk of bias domains relevant to the ITS design. None of the studies reported repeated measures analysis nor tested for trend. However, we re‐analyzed all the ITS studies and overcame this problem. All studies had at least three data points before and after the intervention. Systems for data collection were variable in the studies, but remained the same in all studies during the observation time.

Effects of interventions

1. Primary outcome

Reduction in observed cases of sharps injuries

RCTs

We did not find any randomized controlled trials that addressed this primary outcome.

Cluster‐RCT

The cluster‐RCT (Van der Molen 2011) had an arm of educational intervention alone. Compared to no education, the education alone arm did not decrease the rate of sharps injuries. The outcomes were assessed by questionnaires (RR 0.41, 95% CI 0.14 to 1.21) (Analysis 1.2) and hospital registers (RR 0.46, 95% CI 0.16 to 1.30) (Analysis 1.4). No change in the number of sharps injuries was found either at six or twelve months.

CCTs

Three studies assessed the primary outcome after an educational intervention compared to no education. The follow‐up time was four months in one study (Huang 2002) and two months in two studies (Bijani 2018; Choi 2009). There was a non‐significant decrease in sharps injuries in one study (Huang 2002) and two studies found a significant difference (Bijani 2018; Choi 2009). In the meta‐analysis of these three studies, there was a significant decrease in sharps injuries (RR 0.68, 95% CI 0.48 to 0.95). The summary of the results is presented in Analysis 2.1 and Figure 4.

Figure 4

Forest plot of comparison: 2 Education vs. control: needle stick injuries; CCTs, outcome: 2.1 Sharps injuries, 2 mo

ITS

Table 4, Analysis 3.1, and Analysis 3.2 summarise the ITS results. In the meta‐analysis of two comparable ITS studies, there was a significant level change immediately post‐intervention by 9.3 injuries per 1000 HCWs per year (95% CI ‐14.9 to ‐3.7) (Analysis 3.1, Figure 5). There was a small non‐significant declining trend post‐intervention of 2.3 injuries per 1000 HCWs per year (95% CI ‐7.8 to 12.4) (Analysis 3.2, Figure 6).

Table 4. Effects of interventions of the ITS studies; non‐standardised data

Author, Year	Pre‐intervention level (injuries per 1000 HCWs, 95% CI)	Trend pre‐intervention (estimate, 95% CI)	Level change post‐intervention (estimate, 95% CI)	Trend change post‐intervention (estimate, 95% CI)
El Beltagy 2012, ITS	28.9 (25.4, 32.3)	‐2.8 (‐4.6, ‐0.9)*	‐10.0 (‐15.9, ‐4.0)*	2.0 (‐0.3, 4.2)
Sarbaz 2017, ITS	261.1 (238.6, 283.7)	9.6 (‐2.5, 21.6)	‐77.1 (‐117.1, ‐37.0)*	‐32.5 (‐49.6, ‐15.4)*
Zafar 2009, ITS	44.0 (35.7, 52.3)	0.0 (‐6.5, 6.5)	‐4.5 (‐20.8, 11.8)	‐8.5 (‐17.6, 0.6)

*Indicates statistically significant effects at alpha = 0.05

Figure 5

Forest plot of comparison: 3 ITS, outcome: 3.1 Level change in injury incidence.

Figure 6

Forest plot of comparison: 3 ITS, outcome: 3.2 Trend change in injury incidence

One ITS study, with a ten‐fold incidence and short follow‐up, reported a level change of 77 injuries per 1000 HCWs (95% CI ‐117.1 to ‐37.1) and a declining trend post‐intervention of 32.5 injuries per 1000 HCWs per year (95% CI ‐49.6 to ‐15.4).

As seen in Table 4, one of the studies (Zafar 2009) showed no statistically significant changes in immediate change of level or change in trend, in the annual incidence rate. El Beltagy 2012 reported an immediate reduction in annual incidence rate following the intervention, of 10.0 (95% CI ‐15.9 to ‐4.0) injuries per 1000 HCWs and 39.2 (95% CI ‐58.2 to ‐20.3) injuries per 1000 HCWs, respectively. Sarbaz 2017 only reported eight monthly metrics. It had the highest pre‐intervention level of injuries (261.1 per 1000 HCWs) and reported the highest decreases short‐term by ‐77.1 (95% CI ‐117.1 to ‐37.0) and long‐term ‐32.5 (95% CI ‐49.6 to ‐15.4).

Immediate effects

The meta‐analysis of the three ITS studies (Table 5) showed a non‐significant level change immediately post‐intervention by 20.8 injuries per 1000 HCWs (95% CI ‐43.9 to 2.3). However, for studies with yearly data only, there was a decrease of 9.3 injuries per 1000 HCWs (95% CI ‐14.9 to ‐3.7) (Analysis 3.1).

Table 5. Random‐effects meta‐analysis of the ITS studies

Author, Year

Pre‐intervention level (injuries per 1000 HCWs, 95% CI)

Trend pre‐intervention (estimate, 95% CI)

Level change post‐intervention (estimate, 95% CI)

Trend change post‐intervention

(estimate, 95% CI)

All studies (El Beltagy 2012, ITS; Sarbaz 2017, ITS; Zafar 2009, ITS)

109.7(40.2, 179.3)

‐0.2 (‐5.3, 4.9)

‐20.8 (‐43.9, 2.3)

‐10.8(‐26.2, 4.7)

Original yearly data (El Beltagy 2012, ITS; Zafar 2009, ITS)

35.9 (21.1, 50.7)

‐2.6 (‐4.3, ‐0.8)*

‐9.3 (‐14.9, ‐3.7)*

‐2.3 (‐12.4, 7.8)

*Indicates statistically significant effects at alpha = 0.05.

Effects on trend over time

The meta‐analysis did not show an effect of the intervention on the trend over time (‐2.3, 95% CI ‐12.4 to 7.8) (Table 5, Analysis 3.2).

Reduction in observed cases of splash exposures

Splash exposures were only in one study (Sarbaz 2017) reported separately from sharps injuries, but not in sufficient detail to allow an ITS‐analysis.

Secondary Outcome 1: Observed cases of infections

None of the included studies reported cases of infections.

Secondary Outcome 2: Knowledge related to sharps injuries as measured by surveys to HCWs

Cluster‐RCT

Van der Molen 2011 reported on knowledge and prevalence of needle‐stick injuries; this increased from 31 at baseline to 46 at 12 months post‐intervention as measured by a scale from 0 to 100. However, the difference was not statistically significant compared to control (MD 4.0, 95% CI ‐5.8 to 13) (Analysis 1.5). Attitudes towards safe working with needles were good at baseline and remained the same in the post‐intervention monitoring period. Motivation to safely work with needles amongst HCWs was excellent at baseline and remained the same during the 12 months follow‐up period and there was no room for improvement in this area.

CCTs

There was information on change of knowledge in three CCT studies. Choi 2009 reported on two outcomes related to knowledge on sharps injuries: risk of blood exposure and needle stick injury prevention. Compared to the control group, the knowledge increased for both of these outcomes. Bijani 2018 reported mean awareness scores based on a questionnaire, and Huang 2002 applied a questionnaire on 30 knowledge‐related items. In all these studies, the knowledge outcomes improved significantly after the intervention as compared to the control group (MD 1.82, 95% CI 1.01 to 2.64) (Analysis 2.2).

Secondary Outcome 3: Knowledge related to infectious disease transmission as measured by surveys to HCWs

Cluster‐RCTs:

In Van der Molen 2011, knowledge was reported generally, and knowledge related to infectious disease transmission could not be separated from the data.

CCTs

In all CCTs (Choi 2009, Huang 2002, and Bijani 2018) knowledge was reported generally, and knowledge related to infectious disease transmission could not be separated from the data.

Secondary Outcome 4: Adherence to organizations' occupational health and safety policies and procedures, as observed or reported

Cluster‐RCT

In Van der Molen 2011, facilitation was assessed via the existence of a safety policy, with the intervention group reporting an increased awareness from 38% at baseline to 52% at 12 months follow‐up. Availability of gloves was optimal at baseline and no further room for improvement was seen at follow‐up, while a small decrease in the availability of safety goggles was reported from 83% to 79%. Measures to improve safe work practice with needles saw an increase, whilst improvements in safety culture were also seen.

CCTs

Huang 2002 reported an increase of hepatitis B vaccination among participants from 71.4% pre‐intervention to 91.8% post‐intervention. Knowledge and behavior of the correct management immediately following a needle‐stick injury among nurses who participated in the educational program by Huang 2002 also improved in the following categories; washing the area with soap and water: 10.2% to 91.8%, using a bandage to cover the injury: 12.2% to 85.7%, reporting the injury to the hospital authority: 0% to 69.4%, and seeking post‐exposure counseling and medical assistance: 32.7% to 100%. The participants in Huang 2002 also demonstrated a marked increase in the appropriate handling of a blood spill across three key areas after the intervention: decontaminating with disinfectant: 18.4% to 100%, using protective coverings: 2% to 81.7%, and not using hands to directly clean up a blood spill: 2% to 85.7%.

Secondary Outcome 5: Behavioral changes in clinical practice, as observed or reported

Cluster‐RCT

HCW skills surrounding the safe usage of injection needles were already high and there was only room for minor improvement, while greater improvements were seen in blood tapping needles and infusion needles, as reported by Van der Molen 2011.

CCTs

The CCT studies reported several outcomes related to behavioral changes. Choi 2009 observed small non‐significant changes in the needle and needle box handling (MD 0.89 95% CI 0.11 to 1.67) (Analysis 2.3), in post‐exposure practice (MD 1.36, 95% CI 0.67 to 2.05 ) (Analysis 2.4), and in standard precautions (MD 0.76 95% CI 0.29 to 1.23) (Analysis 2.5). In Huang 2002 and Bijani 2018, the behavioral practices were heterogeneous and difficult to interpret.

ITS

In Zafar 2009, recapping rates noticeably decreased from 13% to 0% post‐intervention, whilst injuries occurring during sampling, cannulation, surgery, and garbage collection remained at a similar percentage before and after the intervention. El Beltagy 2012 reported that post‐intervention injuries while dissembling devices reduced from 6.3% down to 1.9% (p=0.003); injuries from recapping also slightly reduced from 6.6% to 5.6% (NS). Devices being left inappropriately saw a reduction in incidents from 12.4 %to 5.9% (p=0.002), while injuries occurring while placing the devices into sharps disposal containers reduced from 6.3% to 4.5 (NS). Injuries that occurred after disposal as a result of an item sticking out of the disposal container or use of an inappropriate container reduced from 6.0% to 5.0% (NS) incidents. Injuries that happened at other times, such as preparation for reuse, incorrect disposal, and piercing of the sharps disposal container, increased from 9.0% to 15.9% (p=0.009) post‐intervention, as reported by El Beltagy 2012.

Discussion

Summary of main results

We found low‐quality evidence that education and training interventions cause small decreases in the incidence of sharps injuries in the short term after an intervention. Five out of seven included studies showed a significant decrease in the number of sharps injuries. The only cluster‐RCT reported a decreasing (non‐significant) trend in the sharps injuries after the educational intervention. In the two ITS studies reporting yearly data, there were on average 35.9 injuries per year per 1000 HCW, at baseline and the injuries decreased by 9.3 per year per 1000 HCWs. The ITS studies showed that there was an overall decreasing trend in the sharps injuries before the intervention and the educational intervention helped bring down the numbers even further. Only two studies reported splash injuries (Bijani 2018, Sarbaz 2017), and only one study reported them separately from sharps injuries (Sarbaz 2017).

With respect to the secondary outcomes, none of the included studies reported on cases of infection or disease transmission. All included studies reported short‐term changes in knowledge and practice related to sharps injuries, but the level of evidence for these findings was very low‐quality. Three of the included studies (Choi 2009; Huang 2002; Van der Molen 2011) reported on knowledge related to sharps injuries, and infectious disease transmission. Two of the included studies, Huang 2002; Van der Molen 2011, reported on adherence to their organizations' occupational health and safety policies. Other studies (El Beltagy 2012; Huang 2002; Van der Molen 2011; Zafar 2009) reported on behavior changes in clinical practice, including safe needle usage, hand washing, and the use of gloves.

Overall completeness and applicability of evidence

The year of publication for the included studies ranged from 2002 to 2018, with the studies originating from different countries worldwide. Most of the studies came from low‐ or middle‐income countries where the challenges with sharps injuries may be more severe than in the high‐income countries. There is a lack of randomized controlled trials in this area of research, with the majority of studies utilizing an interrupted time series analysis. The follow‐up post‐intervention varied greatly between the studies from two months up to six years.

The reporting of needle‐stick injuries varied across studies, with four studies (El Beltagy 2012; Sarbaz 2017; Van der Molen 2011; Zafar 2009) having official registers at their hospitals to collect injury data, while two studies (Choi 2009; Huang 2002) did not. Unfortunately, it is difficult to obtain country‐specific statistics related to sharps injuries in HCWs due to a lack of national reporting; therefore, identifying worldwide trends in these injuries is not possible.

Since hospital‐wide randomized educational trials may not be feasible, we also included other controlled trials. ITS studies, with hospital injury registers as sources of outcome information, appeared to have a strong design to study educational interventions for sharps injuries. With ITS studies, both short‐term and long‐term trends in sharps injuries could be observed. Also, the use of hospital registers improves the reliability of outcome measurement. We noticed that, in the questionnaire‐based studies, the injury rates were much higher than in the ITS studies. In studies where reporting of injuries is especially required, the threshold for this reporting may be lower than in the register‐based systems. However, both under‐reporting and over‐reporting of sharps injuries are important sources of bias in these studies.

Overall, the studies did not critically appraise the validity of their injury reporting systems. The method of measuring sharps injuries continues to be a challenge. Instead of injuries per person‐year, it has also been suggested to report injuries per operation (Verbeek 2019). One conclusion from this review is that the valid measurement of sharps injuries requires further development.

Some of the included studies presented numerous suggestions for future improvement. El Beltagy 2012 reinforced the importance of education and training as a preventative tool, whilst stating that the introduction of safety‐engineered devices may further reduce injury rates. Huang 2002 advised that future educational interventions should focus on glove usage, sharps injury prevention, and injury reporting and be conducted at least twice per year. All nurses should have access to hepatitis B immunisation and undergo training on universal precautions that future research should investigate the impact of safety‐engineered devices on injury rates. Van der Molen 2011 advised that safety‐engineered devices and interactive communication were effective in reducing injuries. They also recommended that interactive communication include such interventions as training, usage of personal protective equipment, and good housekeeping. Suggested secondary prevention measures consist of hepatitis B vaccination and the management of sharps injuries such as post‐exposure prophylaxis. Zafar 2009 recommended incorporating infection control into the curriculums of medical and nursing schools, including paramedical and technical staff education and training.

Quality of the evidence

We did not identify any randomized controlled trials that assessed education for the prevention of sharps injuries in healthcare settings except for one cluster‐randomized trial (Van der Molen 2011). The methodological problem with this study was the high loss to follow‐up. The level of evidence for the ITS studies was generally low. The ITS studies in our review had typical designs and although they did not report a repeated measures analysis nor tested for trends, we were able to overcome these problems by re‐analyzing the studies.

Our systematic searches make it likely that we found all published studies for this review. A comprehensive search strategy was developed and executed for the purposes of this review. A strength of this review was that we were able to run a comprehensive interrupted time series analysis that was conducted utilizing original data. This provided a precise estimate of the combined results for both pre‐ and post‐intervention.

Potential biases in the review process

We aimed to distinguish outcomes attributable to the educational intervention. In practice, this was not feasible. First of all, the studies applied a variety of educational interventions and, depending on the intervention, the outcomes also varied. A limitation in our review was that we were unable to obtain more information from some of the authors and we were limited by inadequate descriptions of the interventions in the included studies.

Agreements and disagreements with other studies or reviews

Overall, previous reviews have given conclusions on the effects of education on the incidence of sharps injuries. We included fewer studies since it was a requirement that the studies have a controlled design. Our results are in line with other reviews, although more conservative. One review (Tarigan 2015) evaluated the effect of education and safety‐engineered devices (SEDs) on sharps injuries, and discovered the combination of SEDs and education to be most beneficial. Differences exist between our review and this one (Tarigan 2015), which limited the search to studies from 2002 to 2012. We did not limit our literature searches but limited our intervention to education and/or training only and not in combination with other interventions such as engineered devices. The authors also combined different study designs into a single meta‐analysis, while we have undertaken separate meta‐analyses for each study design. However, there were some similarities in that we also excluded studies focused on student HCWs. Our review is similar to another updated Cochrane Review (Reddy 2017), which found very low‐quality evidence on the effectiveness of safe blood collection systems, and safe passive intravenous systems in decreasing rates of sharps injuries.

Figure 1

Study flow diagram

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Figure 4

Forest plot of comparison: 2 Education vs. control: needle stick injuries; CCTs, outcome: 2.1 Sharps injuries, 2 mo

Figure 5

Forest plot of comparison: 3 ITS, outcome: 3.1 Level change in injury incidence.

Figure 6

Forest plot of comparison: 3 ITS, outcome: 3.2 Trend change in injury incidence

Analysis 1.1

Comparison 1: Educational intervention versus no intervention in cluster‐randomized trial, Outcome 1: Needle stick injuries, questionnaires, 6 mo

Analysis 1.2

Comparison 1: Educational intervention versus no intervention in cluster‐randomized trial, Outcome 2: Needle stick injuries, questionnaires, 12 mo

Analysis 1.3

Comparison 1: Educational intervention versus no intervention in cluster‐randomized trial, Outcome 3: Needle stick injuries, hospital registers, 6 mo

Analysis 1.4

Comparison 1: Educational intervention versus no intervention in cluster‐randomized trial, Outcome 4: Needle stick injuries, hospital registers, 12 mo

Analysis 1.5

Comparison 1: Educational intervention versus no intervention in cluster‐randomized trial, Outcome 5: Knowledge (scale 0‐100)

Analysis 2.1

Comparison 2: Educational intervention versus no intervention in CCTs, Outcome 1: Sharps injuries, 2 mo

Analysis 2.2

Comparison 2: Educational intervention versus no intervention in CCTs, Outcome 2: Nurses knowledge of blood exposure

Analysis 2.3

Comparison 2: Educational intervention versus no intervention in CCTs, Outcome 3: Needle and needle box handling

Analysis 2.4

Comparison 2: Educational intervention versus no intervention in CCTs, Outcome 4: Post‐exposure practice

Analysis 2.5

Comparison 2: Educational intervention versus no intervention in CCTs, Outcome 5: Standard precaution

Analysis 3.1

Comparison 3: Educational intervention versus no intervention in ITS, Outcome 1: Level change in injury incidence

Analysis 3.2

Comparison 3: Educational intervention versus no intervention in ITS, Outcome 2: Trend change in injury incidence

Summary of findings 1. Educational intervention compared to no intervention in healthcare workers for preventing sharps injuries and splash exposures

Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Educational intervention compared to no intervention in healthcare workers in cluster RCT
Patient or population: Healthcare workers Setting: Healthcare organisations Intervention: Educational intervention Comparison: No intervention
Outcomes	Risk with no intervention	Risk with Educational intervention	Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Sharps injuries, follow‐up 2 mo (CCT)	463 per 1 000	315 per 1 000 (222 to 440)	RR 0.68 (0.48 to 0.95)	395 (3 Controlled Clinical Trials)	⊕⊝⊝⊝ VERY LOW ^{1 2}
Needle stick injuries, questionnaires, 6 mo (RCT)	140 per 1 000	74 per 1 000 (29 to 188)	RR 0.53 (0.21 to 1.35)	167 (1 cluster‐RCT)	⊕⊕⊝⊝ LOW ^{1 2}
Needle stick injuries, hospital registers, 6 mo (RCT)	38 per 1 000	34 per 1 000 (14 to 83)	RR 0.91 (0.38 to 2.20)	529 (1 cluster‐RCT)	⊕⊕⊝⊝ LOW ³
Needle stick injuries, questionnaires, 12 mo (RCT)	119 per 1 000	49 per 1 000 (17 to 144)	RR 0.41 (0.14 to 1.21)	168 (1 cluster‐RCT)	⊕⊕⊝⊝ LOW ^{1 2}
Needle stick injuries, hospital registers, 12 mo (RCT)	41 per 1 000	19 per 1 000 (7 to 54)	RR 0.46 (0.16 to 1.30)	529 (1 cluster‐RCT)	⊕⊕⊝⊝ LOW ³
Injury rate (ITS), change immediately after intervention, low injury rate	The mean level of injury rates before was 35.9	MD 9.4 lower (14.9lower to 3.8 higher)	‐	2104 (2 Interrupted time series)	⊕⊝⊝⊝ VERY LOW ^{4 5}
Injury rate (ITS), change immediately after intervention, high injury rate	The mean level of injury rates before was 261.1	MD 77.1 lower (117.2 lower to 37.1 higher)	‐	255 (1 Interrupted time series)	⊕⊝⊝⊝ VERY LOW ^{4 5}
Injury rate (ITS), change in trend before and after intervention, low injury rate	The mean trend in injury rates before was ‐10.8	MD 2.3 lower (12.4 lower to 7.8 higher)	‐	2104 (2 Interrupted time series)	⊕⊝⊝⊝ VERY LOW ^{4 5}
Injury rate (ITS), change in trend before and after intervention, high injury rate	The mean trend in injury rates before was 9.6	MD 32.5 lower (49.6 lower to 15.4 higher)	‐	255 (1 Interrupted time series)	⊕⊝⊝⊝ VERY LOW ^{4 5}
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk ratio; MD: Mean Difference
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹ We rated down one level because of imprecision based on wide confidence intervals. ² We rated down with two levels because of risk of bias due to self‐reporting and missing data ³ We rated down with two levels because of wide confidence intervals. ⁴ We rated down one level because of risk of bias due to missing period for data points and few data points. ⁵ We rated down one level because of risk of bias due to a decreasing pre‐intervention trend of injuries in all studies (intervention effect was difficult to detect).

Summary of findings 1. Educational intervention compared to no intervention in healthcare workers for preventing sharps injuries and splash exposures

Table 1. Interventions from included studies

Author, Year, Study Design	Educational Presentation	Interactive Demonstration	Marketing Tools	Web‐Based Education
Bijani 2018, CCT	X	X
Choi 2009, CCT				X
El Beltagy 2012, ITS	X	X	X
Huang 2002, CCT	X	X	X
Sarbaz 2017, ITS				X
Van der Molen 2011, cluster‐RCT	X	X
Zafar 2009, ITS	X	X	X

Table 1. Interventions from included studies

Table 2. Level of participant engagement in training all studies

Author, Year, Study Design	Low Level of Engagement	Medium Level of Engagement	High Level of Engagement	Unable to Identify
Bijani 2018, CCT		X
Choi 2009, CCT		X
El Beltagy 2012, ITS			X
Huang 2002, CCT		X
Sarbaz 2017, ITS				X
Van der Molen 2011, cluster‐RCT		X
Zafar 2009, ITS		X

Table 2. Level of participant engagement in training all studies

Table 3. Risk of bias in the ITS studies

Author, Year, Study Design

Intervention independent from other changes

Sufficient data points

Test for trend

Intervention did not affect data collection

Blinded outcome assessment

Complete data set

Reliable outcome measures

Total

El Beltagy 2012, ITS

Done (1)

This was an

educational intervention only.

Done (1)

Four data points before and five data

points after the intervention

Done (1)

We re‐analyzed

the study for trend.

Done (1)

Data collection remained the same pre‐

and post‐

intervention.

Not clear (0)

The reporting of NSI was done by a net service and the protocol remained the same.

Done (1)

All hospital employees were included in the study.

Done (1)

Hospital had a sharp injury surveillance system prior to and after the intervention. Although not ideal as the possibility of under‐reporting, but appropriate for the study outcome.

Sarbaz 2017, ITS

Done (1)

An education

alone intervention

Done (1)

Four data points

before and four data points after the intervention

Done (1)

We re‐analyzed the study for trend.

Done (1)

Data collection remained the same pre‐ and post‐

intervention.

Done (1)

Infection personnel

and staff were blinded to the study.

Done (1)

All hospital employees were included in the study.

Done (1)

Hospital had a sharp injury surveillance system prior to and after the intervention. Although not ideal as the possibility of under‐reporting, but appropriate for the study outcome.

Zafar 2009

Zafar 2009, ITS

Done (1)

An education

alone intervention

Done (1)

Three data points before and three data points after the intervention

Done (1)

We re‐analyzed the study for

trend.

Done (1)

Comment:

Data collection remained the same

pre‐ and post‐

intervention.

Not clear (0)

Authors did not provide information on blinding.

Done (1)

All hospital employees were included in the study.

Done (1)

Hospital had a sharp injury surveillance system prior to and after the intervention. Although not ideal as the possibility of under‐reporting, but appropriate for the study outcome.

NSI: needle‐stick injury

Table 3. Risk of bias in the ITS studies

Table 4. Effects of interventions of the ITS studies; non‐standardised data

Author, Year	Pre‐intervention level (injuries per 1000 HCWs, 95% CI)	Trend pre‐intervention (estimate, 95% CI)	Level change post‐intervention (estimate, 95% CI)	Trend change post‐intervention (estimate, 95% CI)
El Beltagy 2012, ITS	28.9 (25.4, 32.3)	‐2.8 (‐4.6, ‐0.9)*	‐10.0 (‐15.9, ‐4.0)*	2.0 (‐0.3, 4.2)
Sarbaz 2017, ITS	261.1 (238.6, 283.7)	9.6 (‐2.5, 21.6)	‐77.1 (‐117.1, ‐37.0)*	‐32.5 (‐49.6, ‐15.4)*
Zafar 2009, ITS	44.0 (35.7, 52.3)	0.0 (‐6.5, 6.5)	‐4.5 (‐20.8, 11.8)	‐8.5 (‐17.6, 0.6)
*Indicates statistically significant effects at alpha = 0.05

Table 4. Effects of interventions of the ITS studies; non‐standardised data

Table 5. Random‐effects meta‐analysis of the ITS studies

Author, Year	Pre‐intervention level (injuries per 1000 HCWs, 95% CI)	Trend pre‐intervention (estimate, 95% CI)	Level change post‐intervention (estimate, 95% CI)	Trend change post‐intervention (estimate, 95% CI)
All studies (El Beltagy 2012, ITS; Sarbaz 2017, ITS; Zafar 2009, ITS)	109.7(40.2, 179.3)	‐0.2 (‐5.3, 4.9)	‐20.8 (‐43.9, 2.3)	‐10.8(‐26.2, 4.7)
Original yearly data (El Beltagy 2012, ITS; Zafar 2009, ITS)	35.9 (21.1, 50.7)	‐2.6 (‐4.3, ‐0.8)*	‐9.3 (‐14.9, ‐3.7)*	‐2.3 (‐12.4, 7.8)
*Indicates statistically significant effects at alpha = 0.05.

Table 5. Random‐effects meta‐analysis of the ITS studies

Comparison 1. Educational intervention versus no intervention in cluster‐randomized trial

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Needle stick injuries, questionnaires, 6 mo Show forest plot	1	167	Risk Ratio (M‐H, Random, 95% CI)	0.53 [0.21, 1.35]

1.2 Needle stick injuries, questionnaires, 12 mo Show forest plot	1	168	Risk Ratio (M‐H, Random, 95% CI)	0.41 [0.14, 1.21]

1.3 Needle stick injuries, hospital registers, 6 mo Show forest plot	1	529	Risk Ratio (M‐H, Fixed, 95% CI)	0.91 [0.38, 2.20]

1.4 Needle stick injuries, hospital registers, 12 mo Show forest plot	1	529	Risk Ratio (M‐H, Fixed, 95% CI)	0.46 [0.16, 1.30]

1.5 Knowledge (scale 0‐100) Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Totals not selected

Comparison 1. Educational intervention versus no intervention in cluster‐randomized trial

Comparison 2. Educational intervention versus no intervention in CCTs

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Sharps injuries, 2 mo Show forest plot	3	395	Risk Ratio (M‐H, Random, 95% CI)	0.68 [0.48, 0.95]

2.2 Nurses knowledge of blood exposure Show forest plot	3	395	Std. Mean Difference (IV, Random, 95% CI)	1.82 [1.01, 2.64]

2.3 Needle and needle box handling Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Totals not selected

2.4 Post‐exposure practice Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Totals not selected

2.5 Standard precaution Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Totals not selected

Comparison 2. Educational intervention versus no intervention in CCTs

Comparison 3. Educational intervention versus no intervention in ITS

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
3.1 Level change in injury incidence Show forest plot	3		Std. Mean Difference (IV, Random, 95% CI)	‐20.85 [‐43.99, 2.29]

3.1.1 Studies with yearly rates	2		Std. Mean Difference (IV, Random, 95% CI)	‐9.35 [‐14.94, ‐3.76]
3.1.2 Studies with monthly rates	1		Std. Mean Difference (IV, Random, 95% CI)	‐77.10 [‐117.15, ‐37.05]
3.2 Trend change in injury incidence Show forest plot	3		Std. Mean Difference (IV, Random, 95% CI)	‐10.77 [‐26.21, 4.67]

3.2.1 Studies with yearly rates	2		Std. Mean Difference (IV, Random, 95% CI)	‐2.29 [‐12.41, 7.83]
3.2.2 Studies with monthly rates	1		Std. Mean Difference (IV, Random, 95% CI)	‐32.50 [‐49.60, ‐15.40]

Comparison 3. Educational intervention versus no intervention in ITS